JP2801182B2 - Image coding method - Google Patents

Description

Description: TECHNICAL FIELD The present invention relates to an image encoding method suitable for encoding multi-valued image data. [Prior Art] Conventionally, it is general to suppress the redundancy (generally called compression) by encoding in consideration of the efficiency when transmitting and storing images. In recent years, with the development of digital image processing technology and device technology, images to be transmitted and stored have shifted from binary to multi-valued images, and the resolution has been increased. As a result, the amount of data has become enormous, and a highly efficient coding technique has been required. However, most coding techniques so far are for binary images, and typical ones are MH, MR, MMR, etc. specified for facsimile, and the coding of multi-valued images is essentially Not suitable for Some techniques for coding multi-valued images have also been reported, including block coding, predictive coding, and orthogonal transform coding. Most of these are intended for television images, and are not suitable for encoding general document images and halftone images. In particular, orthogonal transform coding has conventionally used the property that power is concentrated in a small portion of a sequence. According to this method, it is effective for continuous tone images such as photographs, but when applied to an image having a peculiar edge structure such as a document image or a halftone dot image, remarkable deterioration occurs. Also, in orthogonal transform coding, it is known to classify into a plurality of categories according to the direction of an edge in a block and perform coding suitable for the category. However, the conventional method has room for improvement in the accuracy of category classification. <Purpose> In view of the above, an object of the present invention is to provide an image encoding method that can classify categories with higher accuracy than before and that can perform efficient encoding according to the classification result. I do. <Structure for Achieving the Object> To achieve the above object, the image encoding method of the present invention cuts out image information into mXn-sized blocks, performs orthogonal transformation on the blocks, and obtains the obtained orthogonal transformation coefficients. Among the combinations, a plurality of combinations of absolute values of a plurality of orthogonal transform coefficients representing an AC component are obtained, each of which corresponds to a different feature of a block, and a comparison relationship between the plurality of combinations is obtained. Among the combinations, the block is classified into a predetermined category according to a magnitude relationship with a predetermined value corresponding to the predetermined combination,
A predetermined coefficient of the obtained orthogonal transform coefficient is encoded for each category. [Embodiment] In the coding method according to the present invention, multi-valued image information to be coded is cut out into mXn blocks, and this block is used as one unit, which is subjected to orthogonal transformation (Hadamard transformation) to extract features in the block. And categorize them into several categories. Then, optimal quantization is performed for each category. FIG. 1 shows a configuration example of an encoding circuit to which the present invention is applied. 101 is a Hadamard transform unit, 102 is a segmentation unit for performing category classification, and 103 is a selector for selecting a quantizer corresponding to a category based on an output from the segmentation unit 102. 104 is a scalar quantizer for factor Y ₁₁ which represents the brightness average value, 105, 106 and 107 scalar quantizer which performs scalar quantization suitable for each category, 108, 109 and 110 vector quantization suitable for each category It is a vessel.
The number of scalar quantizers and vector quantizers is increased or decreased according to the number of categories. Reference numeral 111 denotes a selector for selecting, from the outputs of the vector quantizers of each category, the output corresponding to the block to be processed, based on the output from the segmentation unit 102. It is what you do. Here, the operation will be described. 8 bits per pixel (256
When the data cut out pixel data into 4-pixel × 4 pixels block gradation) Hadamard converted by Hadamard transform unit 101, the coefficient Y ₁₁ is 0 to 1020, i.e., a value of 10bit, also, Y ₁₂ to Y Each of ₄₄ has a value of -510 to +510, that is, a value of 10 bits. Accordingly, quantized coefficients Y ₁₁ to 8bit data in scalar quantizer 104, which is referred to as intra-block brightness average.
The Y ₁₁ to Y ₄₄ are segmentation over Chillon section 102, (and 16 kinds in this embodiment) which features are classified into several categories according to these coefficients of the image data source according to the magnitude of the value. The selectors 103 and 111 are each provided with a segmentation section 102.
A quantizer corresponding to the category determined in step is selected, and the quantization result is output as a 9-bit code (pattern code). Hereinafter, the Hadamard transform, the segmentation, the scalar quantization, and the vector quantization will be individually described. FIG. 2 shows a 4 × 4 Hadamard transform in the Hadamard transform unit 101, where X is a signal before the conversion and Y is a signal after the conversion. This conversion is Is represented by here given that, Becomes FIG. 3 shows an example of the Hadamard transformation. When there is a vertical edge in the block as shown in (1) in FIG.
_A large value appears at ₁₂ . The sign is the gradient of the brightness of X, that is,-(minus) when the left half is 0 and the right half is 255, and + (plus) when the left half is 255 and the right half is 0 (zero). However, the coefficient Y ₁₁ is for representing the brightness average, not specifically mentioned here. Similarly (2) If the lateral edge is within block, a large value appears in the coefficient Y _21. Others (3) to (16) show patterns of vertical and horizontal lines, key shapes, oblique edges, and oblique lines as shown in the figure. As described above, the image pattern and brightness of the block to be encoded can be known from the result of the Hadamard transform. FIG. 4 shows the meaning of parameters when segmentation is performed on each block in the segmentation section 102. Each parameter is obtained from the Hadamard transform result Y by the following equation. _{VEE = | Y 12 | + |} Y 13 | VLE = | Y 14 | + | Y 24 | HEE = | Y 21 | + | Y 31 | HLE = | Y 41 | + | Y 42 | OTH = | Y 22 | + | Y ₃₃ | + | Y ₄₄ | EF = | VEE−HEE | LF = | VLE−HLE | The parameters are determined by paying attention to the features of the pattern as shown in FIG. VEE is a vertical edge component parameter, VLE is a vertical line component parameter, HEE is a horizontal edge component parameter, HLE is a horizontal line component parameter, OTH is a component parameter of other oblique edges, etc., and EF is a parameter representing the strength of the edge, which is large. It turns out that it is a strong edge. Similarly, LF is a parameter indicating the strength of the line segment. FIG. 5 shows how the blocks are classified into 16 categories by the above-described five parameters in the segmentation section 102, and each branch and each category have the following meaning. This is called a segmentation. 20: Diagonal edge, line 21: Diagonal edge, line + complex pattern 22: Vertical edge + complex pattern 23: Horizontal edge + complex pattern 24: Vertical line + complex pattern 25: horizontal line + complex pattern 30: monotonous flat part 31: flat 32: Vertical edge with small gradation difference 33: Horizontal edge with small gradation difference 40: Vertical edge with large gradation difference 41: Horizontal edge with large gradation difference 50: Vertical edge pattern with large gradation difference 51: Horizontal edge system pattern 52 having a large gradation difference 52: Vertical line system pattern 53 having a large gradation difference 53: Horizontal line system pattern having a large gradation difference Each branch in FIG. 5 has the following meaning. <Branch> Separation of strong diagonal line (large OTH) <Branch> Separation of edge-type pattern (VEE, HEE is large) and linear pattern (VLE, HL large) <Branch> Strong lateral edge pattern (Large EF) Extraction <branch> Separation of vertical edge (large VEE) and horizontal edge (large HEE) <branch> Extraction of complex pattern including diagonal edge (large OTH) <branch> Weak vertical / horizontal pattern (EF is slightly large) Separation of flat part <Branch> Separation of vertical edge (Large VEE) and horizontal edge (Large HEE) <Branch> Extract complex pattern with large OTH <Branch> Complex pattern with large HLE <Branch> Extract a pattern with a large VLE from complex patterns <branch> Separate HEE and VEE from a complex pattern <branch> Extract strong vertical and horizontal line patterns (large LF) <branch> vertical line pattern ( VLE is large) and horizontal line putter (Branch is large) <branch> Extracts a pattern containing edge components (EF is slightly large) from the line system pattern separated in branch 2 <branch> Vertical edge (VEE is large) and horizontal edge ( HEE
<Branch> Separation of a complex pattern (having a large OTH) including a diagonal edge and a flat portion FIG. 6 is a simple explanation of quantization. In the figure, one pattern of the category 40 is described as an example. First, the pattern that has been categorization by segmentation over Chillon unit 102, respectively which were separated for each coefficient, except the Y ₁₁ +/- sign and its absolute value "phase component" and "amplitude component" Call. The amplitude component is weighted to each coefficient, and subjected to nonlinear scalar quantization by scalar quantizers 105-107. Note that the coefficients in the shaded area in the figure are places to be ignored which are predetermined for the category 40, and the numbers are the results of rounding by scalar quantization. FIG. 7 shows the results of non-linear scalar quantization of all 16 categories, and the coefficients indicated by points in the figure are coefficients to be ignored according to the above-mentioned categories. The number following the category number indicates the number of bits assigned to each coefficient. That is, when an image is reproduced from encoded data, coefficients important for reproducing an image close to the original image are extracted for each category, and scalar quantization is executed accordingly. Therefore, by setting the coefficient used for the rounding process of the scalar quantization according to the category, the scalar quantization suitable for the category is performed, and a good image can be reproduced. FIG. 8 is a hardware block diagram for realizing the Hadamard transform in the Hadamard transformer 101, in which all coefficients are calculated in parallel for speeding up. Reference numeral 801 denotes a data input line, and data of 8 bits per pixel is 16 bits.
I will be input. Reference numeral 802 denotes a block buffer for this, which is composed of 4 × 4 = 16 latches. Reference numeral 803 denotes an address generator, which designates a pixel in the X block buffer 802, and The element in the element generator 804 is designated. That is,
Hadamard transform shown earlier It is used to specify the h _kl and x _ij in the calculation. Reference numeral 804 is for generating the above-mentioned _hij , and is constituted by a ROM. The output is h _1l , h _2l , h _3l ... h
_16l and has become a sixteen parallel, 4bit output from the address diene Nereta is input to the address of the ROM, l
= 1,..., Up to 16 will be specified. In the figure, i, j
= 1,..., 4, k, l = 1,. 805, as seen in calculations shown in there destination calculator elements Y ₁₁ coefficients of Y the matrix consists of an adder. 806 and 807 are each arithmetic unit coefficients other than Y _11, as shown in equation constituted by the adder-subtracter. 808,809,
Numeral 810 denotes a 1/4 divider for cutting off lower 2 bits. 811, 812, 813 are the results of these calculations, and Y ₁₁
Is a 10-bit positive number, otherwise a flag bit indicating +/- sign and a 9-bit complement are output. As described above, the circuit operates in parallel with 16 coefficients. The content of the calculation is as shown in the equation. That is, block buffer 80
From 2, data is read for each pixel, and addition or subtraction is performed according to the +/- sign output from the Hkl element generator 804. FIG. 9 is a hardware block diagram of the segmentation section 102 for performing segmentation. 901,902,9
Numerals 03,904,905 denote arithmetic units of parameters for segmentation, each comprising a circuit for converting a complement number into an absolute value (an inverter and an adder) and an adder for adding two or three absolute values. Similarly, 906 and 907 are operation units for each parameter for performing the segmentation, and are configured by subtraction. Reference numerals 908, 909, 910, 911, 912, and 913 perform comparisons as shown in the figure in order to determine categories, and are composed of comparators. When this output is input to the segmentation look-up table ROM918, first, the lower 4 bits of the ROM output a result indicating which of HEE, VEE, HLE, and VLE has the largest value. . Judgment unit
At 917, according to this output, the parameter indicated by the parameter is compared with OTH, and the result is input to the segmentation look-up table ROM 918 again. 914, 915, 916 also perform comparisons as shown in the figure, and are composed of comparators and selectors. The above results are input to the segmentation look-up table ROM918, and the segmentation is performed.
The output category of ROM918 is 4 bits code for the resulting category
Output from 4bit. The determination method is as shown in Table 1. Note that the scalar quantization and vector quantization shown in the present embodiment are not particularly limited, and will not be described in detail. In this embodiment, the Hadamard transform is used because of the easiness of calculation and hardware, but the same concept may be applied using an orthogonal transform (discrete COS transform, slant transform, etc.) similar to this. . Also, in this embodiment, VEE, VLE,
Although the five parameters of HEE, HLE, and OTH are used, basically, if the method of classifying by focusing on vertical and horizontal edges, vertical and horizontal lines, and other diagonal edges, etc., it is possible to replace them. . For example _{UEE = | Y 12 |, YLE} = | Y 14 |, HEE = |
_{Y 21 |, HLE = | Y41} |, OTH = | , and the like | Y _22. In this embodiment, scalar quantization + vector quantization is used for quantization, but the present invention is not limited to this, and scalar quantization or vector quantization alone may be used. Also, "phase component"
We use the method of separating and quantizing into "amplitude components".
It is not limited to this. The size of the matrix as a unit of encoding is not limited to 4 × 4, but can be modified according to the content, resolution, and circuit elements of the image to be encoded. As described above, by focusing on the features in the block, classifying and quantizing, it is possible to execute encoding suitable for the image to be encoded, and to reproduce an image with little deterioration during decoding Becomes It is also possible to determine the area of an image by looking at the classified categories and the categories of adjacent blocks. Also, the shape of the edge mainly depends on the sign, and its gradation level depends on the absolute value. Therefore, when quantizing, the block is separated into a sign (phase component) and an absolute value (amplitude component), so that It is possible to prevent a significant deterioration of the shape of the edge inside. <Effects of the Invention> As described above, according to the present invention, image information is converted to mXn
Cut out into blocks of a size, apply orthogonal transform to the block, and among the obtained orthogonal transform coefficients, are combinations of a plurality of absolute values of a plurality of orthogonal transform coefficients representing AC components, each of which is a different feature of the block Is obtained, and the blocks are classified into a predetermined category according to a comparison relationship between the plurality of combinations and a magnitude relationship between a predetermined value corresponding to a predetermined combination among the plurality of combinations. Since the predetermined coefficients of the obtained orthogonal transform coefficients are encoded for each category, the categories can be separated with higher accuracy than in the past, and the encoding can be performed efficiently.

BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a diagram showing a configuration example of an encoding circuit to which the present invention is applied, FIG. 2 is a diagram illustrating a Hadamard transform, FIG. 3 is a diagram showing a Hadamard transform example, FIG. 4 is an explanatory diagram of the segmentation, FIG. 5 is a diagram showing a specific example of the segmentation, FIG. 6 is a diagram showing an example of quantization, FIG. 7 is a diagram showing a scalar quantization result, and FIG. The figure is a hardware block diagram for realizing the Hadamard transform, and FIG. 9 is a hardware block diagram for realizing the segmentation. In the figure, 101 is a Hadamard transformer, 102 is a segmentation unit, 103 is a selector, and 105 to 107 are scalar quantizers.

Continuation of front page       (56) References IEEE Transaction               on Communicatons V               ol. COM-25 [11] (1977) p.               1285-1292                 Journal of the Institute of Television Engineers of Japan, 39 [10               (1985) Miyahara et al. 898−904                 IEEE Transaction               on Communicatons V               ol. COM-23 [7] (1975-7)               P. 785-786

Claims (1)

(57) [Claims] The image information is cut out into mXn size blocks, the blocks are subjected to orthogonal transformation, and among the obtained orthogonal transformation coefficients, a plurality of combinations of absolute values of a plurality of orthogonal transformation coefficients representing an AC component,
According to a comparison relationship between the plurality of combinations and a magnitude relationship between a predetermined value corresponding to a predetermined combination among the plurality of combinations, a block corresponding to each of the combinations corresponding to different characteristics of the block is obtained. Are classified into a predetermined category, and a predetermined coefficient of the obtained orthogonal transform coefficient is coded for each category.